SLM Lab: A Comprehensive Benchmark and Modular Software Framework for Reproducible Deep Reinforcement Learning

TL;DR

SLM Lab is a modular, reproducible software framework that benchmarks various RL algorithms, enabling fair comparison and introducing new algorithm variants and training methods.

Contribution

It provides a comprehensive, modular RL benchmarking platform with reproducibility, and introduces new algorithm variants and training techniques.

Findings

Modular design ensures fair comparison of RL algorithms.

Introduces a discrete-action variant of Soft Actor-Critic.

Evaluates hybrid synchronous/asynchronous training methods.

Abstract

We introduce SLM Lab, a software framework for reproducible reinforcement learning (RL) research. SLM Lab implements a number of popular RL algorithms, provides synchronous and asynchronous parallel experiment execution, hyperparameter search, and result analysis. RL algorithms in SLM Lab are implemented in a modular way such that differences in algorithm performance can be confidently ascribed to differences between algorithms, not between implementations. In this work we present the design choices behind SLM Lab and use it to produce a comprehensive single-codebase RL algorithm benchmark. In addition, as a consequence of SLM Lab's modular design, we introduce and evaluate a discrete-action variant of the Soft Actor-Critic algorithm (Haarnoja et al., 2018) and a hybrid synchronous/asynchronous training method for RL agents.